New method lets optimization work on twisted problems by re-aligning the coordinates.

Finding the absolute best solution to complex math problems gets harder when you add many variables. Consensus-based optimization, or CBO, uses a group of interacting particles to explore and find this best point. Researchers often add anisotropic noise—random exploration along coordinate axes—to help it handle many variables. However, this trick fails if the problem is rotated even slightly, because it depends on the variables being cleanly separable. To fix this, the authors studied functions that become separable only after a linear coordinate shift. They built a reparametrisation method that provably discovers this rotated coordinate system, letting CBO work in the new variables. This approach also acts as noise adaptation, yielding significant performance improvements on tough benchmark tests.

Why it matters

Scientists and engineers using optimization for complex, high-dimensional problems often face slowdowns when variables aren't independent. This method makes optimization more reliable when problems are rotated, saving computational time in real-world applications.

Questions answered